Journal Pre-proof

A multi-horizon fully coupled thermo-mechanical peridynamics

Changyi Yang , Fan Zhu , Jidong Zhao

PII: S0022-5096(24)00224-2
DOI: https://doi.org/10.1016/j.jmps.2024.105758
Reference: MPS 105758

To appear in: Journal of the Mechanics and Physics of Solids

Received date: 3 April 2024

Revised date: 8 June 2024
Accepted date: 27 June 2024

Please cite this article as: Changyi Yang , Fan Zhu , Jidong Zhao , A multi-horizon fully coupled
thermo-mechanical peridynamics, Journal of the Mechanics and Physics of Solids (2024), doi:
https://doi.org/10.1016/j.jmps.2024.105758

This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition
of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of
record. This version will undergo additional copyediting, typesetting and review before it is published
in its final form, but we are providing this version to give early visibility of the article. Please note that,
during the production process, errors may be discovered which could affect the content, and all legal
disclaimers that apply to the journal pertain.

© 2024 Published by Elsevier Ltd.

 A multi-horizon fully coupled thermo-mechanical peridynamics

Changyi Yang1, Fan Zhu2, Jidong Zhao1,3*

1
Department of Civil and Environmental Engineering, The Hong Kong University of Science and Technology,
HKSAR, China
2 Department of Urban Management, Kyoto University, Kyoto, Japan
3
State Key Laboratory of Mountain Hazards and Engineering Resilience, Chengdu, Sichuan, China
* Corresponding author, jzhao@ust.hk

Abstract
This paper presents a fully coupled thermo-mechanical peridynamic model for simulating
interactive thermo-mechanical material responses and thermally induced fracturing of solids. A
temperature-dependent constitutive model and a deformation-dependent heat conduction model
are derived for state-based peridynamic formulation. The dispersion relation and truncation error
of the state-based peridynamic heat equation are analyzed for the first time. It is found that as non-
locality becoming more pronounced, the dissipative rate of heat is reduced, and the truncation error
becomes larger. A small horizon can effectively mitigate oscillation while reducing the error in the
temperature field. For coupled thermo-mechanical modeling, a novel multi-horizon scheme is
introduced where the thermal field is solved with a different horizon than that of the mechanical
field. The multi-horizon scheme allows for the implementation of a distinct degree of non-locality
for different physical field. Comparing with the constant-horizon scheme, we demonstrate through
numerical examples that the multi-horizon scheme offers smoother and more accurate solutions
and serves a promising option for peridynamics-based multi-physics simulations.

Keywords: Multi-physics, peridynamics, coupling, thermal fracture, computational modeling.

1 Introduction

The peridynamic (PD) theory (Silling, 2000; Silling et al., 2007) is a non-local extension of the
classical continuum mechanics. It is based on integral-differentiation governing equations which
make it inherently suitable for modeling discontinuities. The method has gained increasing
popularity in modeling fracturing in solids (Gao & Oterkus, 2019; Liu et al., 2024; Shi et al., 2022;
Wan et al., 2020; Yang et al., 2024; Zhu & Zhao, 2019a, 2019b). Nonetheless, fracturing in many
natural and industrial processes is accompanied by multiphysics processes. Thermal fracturing is
one of the examples. It is regarded as one of the weathering mechanisms of rock (Buckman et al.,
2021) and is also one of the most devasting defects in addictive manufacturing (Ruan et al., 2023).
In geothermal energy exploitation, cool water is often injected into hot rock to build large scale
crack networks to improve the heat extract efficiency (Xue et al., 2023). The thermal fracturing is
not an independent process but is fully coupled with the mechanical response of material. It is well
known that mutual influence of the temperature variation and material deformation exists.

Given the advantages of the PD theory in modeling discontinuities, it is natural to develop a

coupled thermo-mechanical (TM) PD for modeling the interacting thermal and mechanical fields
associated with the fracturing process. Pioneering works on TM PD include Kilic & Madenci
(2010), Agwai (2011), and Oterkus et al. (2014a, 2014b), who developed PD heat conduction
equation and subsequently fully coupled PD thermomechanics based on the irreversible
thermodynamics. Both the equation of motion and thermal diffusion are formulated with bond-
based PD. More recently, the thermomechanical framework is extended to ordinary state-based
(Zhang & Qiao, 2018, 2020) and non-ordinary state-based (Sun et al., 2023) ones. However, the
heat equation is not explicitly included in these analyses, and only the one-way effects of
temperature on stress are considered. With these coupled thermomechanical methods, PD has been
employed to simulate thermal cracking of various brittle materials including rocks, concretes, ice
and ceramics (Wang et al., 2018, 2019; Bazazzadeh et al., 2020; Chen et al., 2021; Song et al.,
2022; Sun et al., 2023; Song et al., 2021; Zhang et al., 2023). It is worth noting that PD is a versatile
computational framework that can integrate multiple physical fields. Examples include coupled
hydro-mechanical/thermo-hydro-mechanical modeling for porous media (Oterkus et al., 2017;
Song & Silling, 2020; Menon & Song, 2021, 2022, 2023; Ni et al., 2022, 2023), fluid-driven
fracturing (Sun et al., 2022a; Sun et al., 2022b; Yang et al., 2024), and ion diffusion and chemical
reactions (Chen & Bobaru, 2015; Wang et al., 2018; Wu & Chen, 2023; Yang et al., 2020).
The PD theory is known to be a non-local theory, meaning that a material point interacts not
only with its immediate neighboring points but also points at a finite distance, namely the horizon.
This non-locality enables PD to well capture the crack propagation process in fracturing problems.
However, when incorporating different physical fields except the mechanical field in the PD
2
framework, it is important to realize that the non-local nature of PD may become a possible source
of error, since many physical processes are governed by local equilibrium. For heat conduction
problems, for example, the transfer of heat occurs through direct contact between adjacent
materials and such process is described by the local governing equations in classical
thermodynamics. However, when simulating either pure heat conduction or couple thermo-
mechanical problems in PD, the heat transfer is still modeled within a finite zone of influence.
Hence, the point-skipping heat transfer contradicts the physics underlying actual conduction when
observed from a macroscopic perspective. The parametric studies conducted by Agwai (2011)
reveal that the temperature distribution significantly deviates from the analytical solution as the
horizon becomes larger. Moreover, the PD formulation is found to be dispersive. When modeling
mechanical responses of materials, dispersion occurs for high frequency waves mainly owing to
the non-local nature of PD and it may also be attributed to discretization (Bessa et al., 2014; Bažant
et al., 2016; Butt et al., 2017). When modeling fluid flow in porous media using PD, Ni et al.,
(2022) analyzed the length scale of different phases and mentioned that dispersive behavior may
occur due to the Laplacian term in Darcy’s law. In view of the similarity between Fourier’s law
and Darcy’s law, when applying PD for modeling the thermal field, numerical issues such as
oscillation of field variables and dispersion of thermal waves can be expected. However, to the
best knowledge of the authors, no prior investigation has been performed to address dispersion
characters and error mitigation for PD formulation of the heat conduction equation.
For the thermo-mechanical model within pure PD framework, most of the existing methods
are developed within the framework of bond-based PD, which assumes that bonds behave
independently from each other. Potential drawbacks of the bond-based PD have been discussed in
detail by Silling et al. (2007) including limitation on the Poisson’s ratio of materials and difficulties
in incorporating complex constitutive behavior of materials. When modeling heat conduction by
bond-based PD, a micro-conductivity is always needed to make analogy between classical material
conductivity and peridynamic material conductivity. However, the determination of this parameter
itself is non-trivial since it depends on the weight function. Either complicated analytical
expression or numerical calculation of this parameter is needed. These drawbacks associated with
the bond-based PD can be greatly eased by adopting the state-based PD. Notably, Ni et al. (2023)
recently employed the state-based PD to model mechanical response of the solid phase and finite
element method (FEM) to model heat conduction and fluid flow in porous media, with additional
coupling techniques introduced between PD and FEM.

The objective of this work is to present a fully coupled thermo-mechanical formulation suited
for the state-based PD, with rigorous study of its dispersion characters and discretization-induced
error. For the coupling between the thermal and mechanical fields, we show derivation of a

3
temperature-dependent elastic model together with a deformation-dependent heat conduction
model. For heat conduction problem, the dispersion characters of the formulation is analyzed
through spectral approach and the truncation error is studied by Taylor expansion of the discretized
governing equation. It will be shown that the state-based TM PD formulation may experience
amplified numerical oscillation and errors owning to its non-local nature. As a strategy of remedy,
a novel multi-horizon computational scheme is introduced for the coupled thermo-mechanical PD
model. It is demonstrated that the multi-horizon scheme offers appreciable improvements in the
simulation accuracy.
This paper is organized as follows. In Section 2, we present an a new fully coupled
thermomechanical state-based PD model, including a temperature-dependent constitutive model
and a heat conduction model with consideration of the effects of deformation on temperature. In
Section 3, the dispersion relation and error analyses of PD heat equation are investigated from a
mathematical perspective. Section 4 presents the principles and algorithm of a multi-horizon
scheme for coupled thermo-mechanical modeling. Multiple numerical examples are provided for
validation purpose in Section 5, including heat conduction model, coupled thermomechanical
models, and thermal fracturing model. Finally, conclusion and a discussion on the proposed
computational scheme are given in Section 6.

2 Fully coupled thermomechanical state-based peridynamics

In the PD theory, a continual media is modeled by discretized material points. The material points
interact with each other within a certain distance range, as illustrated in Fig. 1. This range is
denoted by 𝛿 and is named horizon. The set of material points within the horizon of 𝒙 is termed
family of 𝒙 and is denoted by Ω𝑥 . Under non-isothermal conditions, the interactions between
material points involve not only the mechanical response but also the heat transfer process. As
illustrated in Fig. 1, when a master point, 𝑥, has a hotter neighboring point, 𝒙′, the bond force
applies along with the heat flow from 𝒙′ to 𝑥. On the contrary, when the master point, 𝑥, is colder
than a neighboring point, 𝒙′, the heat flows from 𝑥 to 𝒙′. The material response is determined by
the coupled thermal and mechanical fields. In other words, the total force acting on material point
𝑥 consists of a structural force component, which follows the constitutive law of the material under
isothermal condition, and a thermal component which originates from the temperature change.
Meanwhile, the temperature of material point 𝑥 is determined by the collective heat flux of
neighboring points as well as the deformation of those neighbors. The key of developing a fully
coupled thermomechanical PD model involves a temperature-dependent constitutive model and a

4
deformation-dependent heat conduction model. The two key aspects are addressed in Section 2.1
and Section 2.2. The formulations are derived for the state-based PD.

Hotter

Heat flux Force state Heat flux

Isothermal
(a) (b) (c)

Colder : family of material point x

: range of interactions with material point x

Fig. 1. Schematics of thermomechanical state-based PD. Shown in the figure are interactions between the
master material point and (a) hotter material point (force state and heat flux form 𝒙′ to 𝑥); (b) isothermal
material point (force state only); (c) colder material point (force state and heat flux from 𝑥 to 𝒙′).

2.1 Temperature-dependent constitutive model

The general form of the equation of motion in the state-based PD can be expressed by

𝜌(𝒙)𝒖̈ (𝒙, 𝑡) = ∫ [𝑻〈𝒙′ − 𝒙〉 − 𝑻〈𝒙 − 𝒙′〉] d𝑉𝑥′ + 𝒃(𝒙) (1)

Ω𝑥

where 𝜌 represents density at a material point, u represents material point deformation, b

represents body force density and 𝑉𝑥′ denotes the volume of a neighbor point 𝑥′. 𝑻〈𝒙′ − 𝒙〉 is a
force state which represents the force exerted by 𝒙′ on 𝒙 and the angle bracket represents the bond
that the state operates on. Note that if 𝑻〈𝒙′ − 𝒙〉 equals 𝑻〈𝒙 − 𝒙′〉, the state-based PD reduces to
the bond-based PD. The force state maps the deformation of a bond into a force vector and different
constitutive models can be implemented in the calculation of the force state. For an ordinary
material, the force state act on the bond 𝝃 = 𝒙′ − 𝒙 is computed by (Silling et al., 2007)
𝒀
𝑻〈𝒙′ − 𝒙〉 = 𝑡 (2)
‖𝒀‖

where 𝑡 represents a scalar force state and 𝒀 is the deformed bond vector. It is assumed that Eq.
(2) remains applicable under non-isothermal conditions, and therefore the key of developing a
thermomechanical PD solid model comes to determine the expression of 𝑡 under different
temperatures. The derivation of t for the bond-based PD is available in previous work (Oterkus et
al., 2014a) and the present study is focused on deriving t that fits into the state-based PD.

5
 For an elastic material, the scalar force state t can be obtained by taking the Fréchet derivative
of energy density function 𝑊 with respective to the extension state 𝑒 as 𝑡 = ∇𝑒 𝑊. The term ∇𝑒 𝑊
is defined as (Silling et al., 2007)

𝑊(𝑒 + ∆𝑒) = 𝑊(𝑒) + ∇𝑒 𝑊 ∙ ∆𝑒 + 𝑜(‖∆𝑒‖) (3)

in which 𝑒 can be calculated by ‖𝒀‖ − ‖𝝃‖; ∆𝑒 represents an increment in the extension state;
𝑜(‖∆𝑒‖) is the residual error. Note that the non-local energy density function 𝑊PD in PD form
should be equivalent to classical local energy density function 𝑊 (Le et al., 2014), which can be
readily obtained from the definition 𝑊 = ∫ 𝜎𝑖𝑗 d𝜀𝑖𝑗 and the generalized Hooke’s law (Wang, 2017;
Zhang & Qiao, 2018, 2020) as

𝜆 d𝑉 2 d𝑉
𝐺 ∑ 𝜀𝑖𝑗 𝜀𝑖𝑗 + ( ) − 3𝑘𝛽𝛥𝛩 ( ) + 𝑓(Θ) , 3D
2 𝑉 𝑉
𝑖,𝑗=1,2,3
𝑣 2 𝜆 1 − 2𝑣 2 d𝑆 2 1 − 2𝑣 d𝑆
𝑊 = 𝐺 ∑ 𝜀𝑖𝑗 𝜀𝑖𝑗 + [𝐺 ( ) + ( ) ] ( ) − 3𝑘𝛽Δ𝛩 + 𝑓(𝛩), plane stress (4)
1−𝑣 2 1−𝑣 𝑆 1−𝑣 𝑆
𝑖,𝑗=1,2
𝜆 d𝑆 2 d𝑆
𝐺 ∑ 𝜀𝑖𝑗 𝜀𝑖𝑗 + ( ) − 3𝑘𝛽Δ𝛩 + 𝑓(𝛩) , plane strain
2 𝑆 𝑆
{ 𝑖,𝑗=1,2

where 𝐺 and 𝑘 are shear and bulk moduli, respectively; 𝜆 is lame constant and 𝑣 is Poisson’s ratio;
𝜎𝑖𝑗 and 𝜀𝑖𝑗 are stress tensor and strain tensor, respectively; d𝑉/𝑉 and d𝑆/𝑆 represent volumetric
strain and plane strain, respectively; 𝛽 is the linear coefficient of thermal expansion; Δ𝛩 denotes
the temperature change between the current state and a reference state; 𝑓(𝛩) is a potential energy
function associated only with current temperature, which is independent of the stress state.
The non-local peridynamic form of elastic energy density function should converge to the
classical local density function when the horizon reduces to zero, by making an analogy between
the two (Zhang & Qiao, 2020), one can obtain the peridynamic form energy density function as

𝑘 2 15𝐺
𝜃 + (𝑤〈‖𝝃‖〉𝑒 d ) ∙ 𝑒 d − 3𝑘𝛽ΔΘ𝜃 + 𝑓(Θ) , 3D
2 2𝑚
1 ′ 2 4𝐺
𝑊PD = 𝑘 𝜃s + (𝑤〈‖𝝃‖〉𝑒 d ) ∙ 𝑒 d − 2𝑘 ′ 𝛽ΔΘ𝜃s + 𝑓(Θ) , plane stress (5)
2 𝑚
1 ′ 2 4𝐺 d d ′ (1
{2 𝑘 𝜃s + 𝑚 (𝑤〈‖𝝃‖〉𝑒 ) ∙ 𝑒 − 2𝑘 + 𝑣)𝛽ΔΘ𝜃s + 𝑓(Θ) , plane strain

where 𝑒 d represents the deviatoric part of extension state, which can be calculated by 𝑒 − 𝑒 i ; 𝑒 i
represents the isotropic part of extension state and can be related to dilation 𝜃 and plane dilation
𝜃s through

6
 𝜃
‖𝝃‖ ,3D
i
𝑒 ={ 3 (6)
𝜃s
‖𝝃‖ , 2D
2
the weight volume 𝑚, dilation 𝜃 (for 3D) and plane dilation 𝜃s (for 2D) are calculated by

𝑚 = ∫ 𝑤⟨‖𝝃‖⟩ ‖𝝃‖2 d𝑉𝑥′ (7)

Ω𝑥

3
𝜃= ∫ 𝑤⟨‖𝝃‖⟩ ‖𝝃‖ 𝑒 d𝑉𝑥′ (8)
𝑚 Ω𝑥
2
𝜃s = ∫ 𝑤⟨‖𝝃‖⟩ ‖𝝃‖ 𝑒 d𝑉𝑥′ (9)
𝑚 Ω𝑥

Here 𝑤⟨‖𝝃‖⟩ is the weight function that measures the influence of neighboring points. It is chosen
‖𝝃‖ 2
−( )
to be the Gaussian function in the form of 𝜔⟨‖𝝃‖⟩ = 𝑒 𝛼𝛿 where the parameter 𝛼 is selected to
′
be 0.5. For 2D cases, 𝑘 is defined as the plane bulk modulus and can be calculated based on the
Young’s modulus 𝐸 and Poisson’s ratio (𝑣) by
𝐸
, plane stress
′ 2(1 − 𝑣)
𝑘 = (10)
𝐸
, plane strain
{2(1 + 𝑣)(1 − 2𝑣)

Substituting the non-local elastic strain energy density 𝑊PD into Eq. (3) yields the scalar force
state in the form of

3𝑘 15𝐺
(𝜃 − 3𝛽ΔΘ)𝑤〈‖𝝃‖〉‖𝝃‖ + 𝑤〈‖𝝃‖〉𝑒 d , 3D
𝑚 𝑚
2𝑘′ 8𝐺
𝑡= (𝜃s − 2𝛽ΔΘ)𝑤〈‖𝝃‖〉‖𝝃‖ + 𝑤〈‖𝝃‖〉𝑒 d , plane stress (11)
𝑚 𝑚
2𝑘′ 8𝐺 d
[𝜃
{ 𝑚 s − 2(1 + 𝑣)𝛽ΔΘ]𝑤〈‖𝝃‖〉‖𝝃‖ + 𝑚 𝑤〈‖𝝃‖〉𝑒 , plane strain

It can be noted that the first term in the scalar force state represents a volumetric term due to
dilation and temperature change whereas the last term represents a force originated from the
deviatoric bond extension.

7
2.2 Deformation-dependent heat conduction model

The general form of the fully coupled peridynamic heat equation writes (Oterkus et al., 2014a)

D𝛩
𝜌𝑐 = ∫ [𝒉〈𝒙′ − 𝒙〉 − 𝒉〈𝒙 − 𝒙′ 〉 + 𝛩𝑩〈𝒙′ − 𝒙〉 ∙ 𝒀̇〈𝒙′ − 𝒙〉] d𝑉𝑥′ + 𝜌𝛩b (12)
D𝑡 Ω𝑥

in which 𝑐 is the specific heat capacity; 𝛩 and 𝛩b represent the absolute temperature and the
volumetric heat generation per unit mass, respectively; 𝒉〈𝒙′ − 𝒙〉 denotes the heat flux state;
𝑩〈𝒙′ − 𝒙〉 represents the thermal modulus state and 𝒀̇〈𝒙′ − 𝒙〉 represents by rate of change of
extension. The product of these two terms gives the effect of deformation on temperature. However,
in the previous studies, both the heat flow state and the heat flux state are determined for the bond-
based PD only. In the present study, we derive a more comprehensive form of the heat equation
for the state-based PD by transforming the classical heat equation and Fourier’s law using a non-
local differential operator.

For heat conduction problems, the continuity equation in its local form is expressed by

D𝛩
𝜌𝑐 = −∇ ∙ 𝒒 + 𝜌𝛩b (13)
D𝑡

where 𝜌 is the density; 𝑐 is the specific heat capacity; 𝛩 and 𝛩b represent the absolute temperature
and the volumetric heat generation per unit mass, respectively; 𝒒 is the heat flux and can be related
to temperature by the Fourier’s law as

𝒒 = −𝑘h ∇ ⊗ 𝛩 (14)

where 𝑘h represents the heat conductivity. To derive the heat conduction equation in peridynamic
form, we utilize the non-local operators which was introduced in PD to approximate differentiation
through integration. Successful attempts on non-local operators have been made to achieve this
conversion between derivatives and integrals by Taylor expansion such as Madenci et al., (2019),
Rabczuk et al., (2019) and Ren et al. (2020b, 2020a). In this paper, the non-local gradient operator
and divergence operator proposed by Bergel & Li (2016) are adopted as

∇ ⊗ 𝑨 = [∫ 𝑤⟨‖𝝃‖⟩(∆ ∙ 𝑨) ⊗ 𝝃 d𝑉𝑥 ′ ] 𝑴−1

𝑥 (15)
Ω𝑥

8
 ∇ ∙ 𝑨 = ∫ 𝑤⟨‖𝝃‖⟩(∆ ∙ 𝑨) ∙ (𝑴−T
𝑥 𝝃) d𝑉𝑥 ′ (16)
Ω𝑥

where 𝑨 is a vector field and ∆ ∙ 𝑨 = 𝑨𝒙′ − 𝑨𝒙 is defined as the non-local difference operator; the
shape tensor 𝑴𝑥 is calculated as

𝑴𝑥 = ∫ 𝑤⟨‖𝝃‖⟩ 𝝃 ⊗ 𝝃 d𝑉𝑥′ (17)

Ω𝑥

Note that even if 𝑨 reduces to a scalar, such as temperature in Eq. (14), the abovementioned
operators remain applicable. The above non-local operator has been proven to converge to the
local gradient and divergence operators when the horizon approaches zero (Tu & Li, 2017).
Substituting Eqs. (15)-(16) into Eqs. (13)-(14) yields the non-local heat equation and Fourier’s law

D𝛩 −T
𝜌𝑐 = ∫ 𝜔⟨‖𝝃‖⟩( 𝒒𝑥 𝑴−T ′
𝑥 𝝃〈𝒙′ − 𝒙〉 − 𝒒𝑥 ′ 𝑴𝑥 ′ 𝝃〈𝒙 − 𝒙 〉 )d𝑉𝑥 ′ + 𝜌𝛩b (18)
D𝑡 Ω𝑥

𝒒𝒙 = −𝑘h [∫ 𝜔⟨‖𝝃‖⟩ 𝛩〈𝒙′ − 𝒙〉 𝝃 d𝑉𝑥 ′ ] 𝑴𝑥−1 (19)

Ω𝑥

where the temperature state is defined as

𝛩〈𝒙′ − 𝒙〉 = 𝛩𝒙′ − 𝛩𝒙 (20)

The Eqs. (18) and (19) are implemented in the state-based PD. Note that the classical heat
conductivity 𝑘h is directly used in the state-based PD formulation. This eliminates the need of
finding a micro-conductivity value in bond-based PD. However, the local heat equation given in
Eq. (13) does not consider the possible temperature variation induced by deformation. To consider
such effect, an additional term has been added into Eq. (18) and the fully coupled heat conduction
equation take the form of

D𝛩
𝜌𝑐 = ∫ [𝜔⟨‖𝝃‖⟩(𝒒𝒙 𝑴−T −T ̇
𝑥 + 𝒒𝒙′ 𝑴𝑥′ )𝝃 + 𝛩𝑩〈𝒙′ − 𝒙〉 ∙ 𝒀〈𝒙′ − 𝒙〉]d𝑉𝑥 ′ + 𝜌𝛩b (21)
D𝑡 Ω𝑥

where the thermal modulus state 𝑩〈𝒙′ − 𝒙〉 can be determined by referencing to the force state
given in Eq. (11) given that the thermal effect on the force state should be represented by
𝑩〈𝒙′ − 𝒙〉Δ𝛩

9
 3𝑘 𝒀
𝛽𝑤〈‖𝝃‖〉‖𝝃‖ , 3D
𝑚 ‖𝒀‖
4𝑘′ 𝒀
𝑩〈𝒙′ − 𝒙〉 = 𝛽𝑤〈‖𝝃‖〉‖𝝃‖ , plane stress (22)
𝑚 ‖𝒀‖
4𝑘′(1 + 𝑣) 𝒀
𝛽𝑤〈‖𝝃‖〉‖𝝃‖ , plane strain
{ 𝑚 ‖𝒀‖

Remark 1. The effect of deformation on temperature variation may be negligible for typical
brittle-elastic materials owning to the small deformation before fracturing. However, such effect
is more pronounced when a material undergoes plastic deformation. An example refers to growing
fracture in metals where a plastic region forms ahead of the crack tip which leads to temperature
increase with the dissipation of mechanical energy. Nevertheless, when dealing with plastic
material responses with large deformations, it is essential to conduct a careful investigation to
determine if the current total-Lagrangian scheme remains applicable and appropriate.

2.3 Damage model

In peridynamics, the fracture of solid is modeled by the breakage of bonds between material points.
A common approach to determine bond breakage under mechanical loads is through the critical
stretch model (Silling & Askari, 2005), which states that a bond breaks irreversibly when it exceeds
a predefined limit of strain. This critical stretch for bond failure can be expressed as

5𝐺𝑐
𝑠𝑐 = √ (23)
9𝐾𝛿

where 𝐺𝑐 denotes the critical energy release rate. The local damage of a material point can be
calculated by averaging the failure at bond level

∫Ω 𝑔〈𝝃〉 d𝑉𝑥 ′
𝑥
𝜑 = 1− (24)
∫Ω d𝑉𝑥 ′
𝑥

in which 𝑔〈𝝃〉 is a binary scalar-valued function that takes the value of either 1 (for intact bond) or
0 (for broken bond).
In the case of thermally induced fracturing, opening of fractures on the free surface, away from
the heat source, is predominantly attributed to the tensile stress (and subsequently thermal
expansion) induced by temperature change. Our experience shows that the performance of the
critical stretch model in such scenario is less satisfactory than expected. A possible reason is that

10
the critical stretch value is dependent of the weight function, using a weight function different from
the one for derivation of Eq. (23) can potentially induce inaccuracies. Also, since different horizons
are used for the thermal and mechanical fields (as will be introduced in later sections), the critical
stretch of bonds, which is assessed with the horizon of the mechanical field, may deviate from the
length scale that is used for thermal field modeling. Therefore, in this study, an additional tensile
stress-based failure model is implemented alongside the critical stretch model. With this model, a
material point fails when its major principal stress exceeds the tensile strength of the material.
Therefore, the local damage of a material point can reach 1.0 under two circumstances: a) bonds
between all interacting materials break; and b) major principal stress exceeds the material’s tensile
strength.

3 Peridynamic thermal wave dispersion and error analysis

It is well known that the non-locality of peridynamics brings wave dispersion in transient
analysis. The dispersive response of both bond-based and state-based PD in modeling elastic wave
propagation has been extensively investigated based on the spectral approach (Bažant et al., 2016;
Butt et al., 2017; Gu et al., 2016; Kulkarni & Tabarraei, 2018; Mikata, 2012; Silling, 2016). It was
found that both the non-local nature of PD and the discretization introduce dispersion. Furthermore,
the dispersion depends on a variety of model settings including horizon, weight function and
material point size. In general, larger horizon and coarser mesh are associated with stronger
dispersion, while the state-based formulation is more dispersive than the bond-based formulation
(Bažant et al., 2016).

The heat conduction is a dissipative process, which tends to dampen and smoothen out the
waveforms over time. Moreover, thermal waves can exist when variations or fluctuations in
temperature propagates through a medium. For example, in the case of periodic heating or cooling
(e.g., in laser-induced heating or pulsed heating experiments), thermal waves can be generated.
These thermal waves exhibit wave-like behaviors such as dissipation and dispersion just like other
types of waves. When modeling heat conduction processes by PD, due to the non-local feature of
the method, heat transfer is modeled not only between adjacent neighboring points but also
between points that are not immediate neighbors. That is, the heat transfer may skip some
intermediate material points lying between the paired points of a bond. This is true from an atomic
perspective, which however does not hold at macro- or meso-scale where PD is generally applied.
Therefore, it is essential to further examine the dispersion relation and numerical errors of the non-
local PD heat equation, which will offer insights into the non-local effect on the accuracy of the
numerical solution and facilitate finding possible remedies. To the best knowledge of the authors,

11
the study on thermal wave dispersion properties of the state-based PD has not been done in past
studies.

3.1 Dispersion relation

In this section, we analyze dispersion relations for heat conduction in state-based peridynamic
continua. Since the 2D and 3D analyses are similar to 1D case, only the detailed derivations of 1D
case are presented here for clarity. To obtain the dispersion relation, a periodic thermal wave at
material point 𝑥 and time 𝑡 is introduced as

𝛩(𝑥, 𝑡) = 𝐴𝑒 𝑖(𝑘𝑥−𝜔𝑡) (25)

in which 𝐴 is the amplitude of the wave (the maximum temperature from the equilibrium state); 𝑘
and 𝜔 denote the wave number and angular frequency, respectively; 𝑖 represents the imaginary
unit. Note that it is enough to analyze harmonic temperature variations, since any periodic wave
or non-periodic wave can be expanded into Fourier series or Fourier transform, respectively.

Considering the heat conduction in an infinite 1D elastic state-based peridynamic continuum,

the non-local governing equation becomes (neglecting the body heat generation and heat generated
by deformation)

𝛿
D𝛩
𝜌𝑐 = ∫ 𝑤(|𝜉|)(𝑞𝑥 𝑀𝑥−T + 𝑞𝑥′ 𝑀𝑥′
−T )
𝜉 d𝑉𝑥 ′ (26)
D𝑡 −𝛿

where the bond length 𝜉 = 𝑥′ − 𝑥 becomes a scalar value. Assume a unity weight function for
convenience, the shape tensor also degenerates to a scalar value and is the same at all points (except
at the boundary) for the 1D case

𝛿
2𝛿 3
2
𝑀𝑥 = 𝑀𝑥′ = ∫ 𝑤(𝜉)𝜉 d𝜉 = (27)
−𝛿 3

According to Eq. (19), the heat flux at two different material points 𝑥 and 𝑥′ are given as

𝛿
[∫−𝛿(𝛩𝑥+𝜉 − 𝛩𝑥 )𝜉 d𝜉]
𝑞𝑥 = −𝑘h 𝛿
(28)
∫−𝛿 𝑤(𝜉)𝜉 2 d𝜉

12
 𝛿
[∫−𝛿(𝛩𝑥 ′ +𝜂 − 𝛩𝑥′ )𝜂 d𝜂 ]
𝑞𝑥′ = −𝑘h 𝛿
(29)
∫−𝛿 𝑤(𝜂)𝜂2 d𝜂

Substituting the periodic thermal cycling given in Eq. (25) into Eqs. (28-29) and Eq. (26), one
can obtain

9𝑘h 𝛿 𝛿 𝛿
−𝜌𝑐𝜔𝑖 = 6 ∫ [∫ (𝑒 − 1)𝜉d𝜉 + ∫ 𝑒 𝑖𝑘𝜉 (𝑒 𝑖𝑘𝜂 − 1)𝜂d𝜂 ] 𝜉d𝜉
𝑖𝑘𝜉 (30)
4𝛿 −𝛿 −𝛿 −𝛿

This equation can be further simplified by using the Euler formula and the symmetric property of
material points within horizon into the form of

9𝑘h 𝛿 𝛿
𝑖𝑘𝜉
𝛿
−𝜌𝑐𝜔𝑖 = ∫ [∫ 𝑖𝜉sin(𝑘𝜉)d𝜉 + 𝑒 ∫ 𝑖𝜂sin(𝑘𝜂)d𝜂] 𝜉d𝜉 (31)
4𝛿 6 −𝛿 −𝛿 −𝛿

By calculating the integrals in Eq. (31), we obtain the 1D dispersion relation of PD heat equation
in terms of angular frequency

2
9𝑘h sin(𝑘𝛿) − 𝛿𝑘cos(𝑘𝛿)
𝜔 = −𝑖 [ ] (32)
𝜌𝑐 𝑘2𝛿 3

The imaginary solution obtained in Eq. (32) represents the dissipative nature of heat
conduction. This can be demonstrated by substituting Eq. (32) into Eq. (25). By doing so, one can
derive the expression 𝛩(𝑥, 𝑡) = 𝐴𝑒 𝑖𝑘𝑥 𝑒 −|𝜔|𝑡 , in which 𝐴𝑒 𝑖𝑘𝑥 represents the initial wave shape and
the length of 𝜔 indicates the damping rate of thermal wave. This indicates that solving heat
equation by PD non-local operators will not induce extra dispersion and will not violate the nature
of heat conduction. However, the dispersion relation of PD heat equation is clearly different from
the local analytical solution (𝜔 = −𝑖𝑘h 𝑘 2 /𝜌𝑐). A quantitative comparison is presented between
the results of local methods and PD for the normalized angular frequency (𝜔n = 𝜔/−𝑖) plotted
against the square of the wave number in Fig. 2. It can be found that the numerical results converge
to the local analytical solution as horizon approaches zero. Moreover, the dissipative rate of PD
heat equation becomes smaller as the horizon increases, especially for waves with high frequency
and short wavelength. This means the magnitude of thermal wave dampens slower when adopting
larger horizon. The mathematical findings for state-based PD in this study are consistent with the
parametric analyses conducted by Agwai (2011) for bond-based PD, in which a heated bar with

13
zero temperature boundary is modeled. In Section 3.3, we will also conduct a benchmark for 1D
heat conduction problem to showcase the effects of horizon within the state-based PD framework.

100 Local heat equation

d = 0.02
d = 0.1
Normalized frequency, wn : -i

80 d = 0.2
d = 0.3
60 d = 0.4
d = 0.5

40

20

0 20 40 60 80 100
Wave number square, k2

Fig. 2. Relation between normalized angular frequency and wave number for state-based PD heat equation.

Remark 2. The state-based PD heat equation does not alter the inherent non-dispersive nature of
the heat conduction process; however, it significantly reduces the dissipation rate of thermal
waves, i.e., a slower heat conduction process, with increasing horizon.

3.2 Truncation error of discretized PD heat equation

It is widely recognized in the FEM and finite difference method (FDM) that, despite of using
non-dispersive governing equations, the discretization process itself can introduce additional
dispersion (Bažant et al., 2016; Butt et al., 2017). Discretization generates inherent non-local
properties because mass is lumped at discrete elements or points. It is a natural assumption to
expect that the error introduced by discretization can be eliminated when the mesh is refined
sufficiently, such as reaching the atomic size scale. Unfortunately, achieving a material point size
that fine is computationally unfeasible due to the prohibitively high computational cost it would
entail. Like FEM and FDM, the PD heat equation is not exempt from errors arising from
discretization. In this section, a comprehensive analysis is performed to identify and examine the
sources of error in the PD heat equation. The objective is to provide insights and guidance on
minimizing these errors beyond simply refining the mesh.

14
 Consider the discretized form of 1D PD heat equation, where forward difference is utilized to
approximate the temporal derivative and the horizon is meshed into (𝐽 + 1) discretized material
points

𝐽
𝛩𝑖𝑛+1 − 𝛩𝑖𝑛
𝜌𝑐 = ∑ 𝑤(𝜉)(𝑞𝑖 𝑀𝑖−T + 𝑞𝑗 𝑀𝑗−T ) 𝜉 (33)
∆𝑡
𝑗

where 𝛩𝑖𝑛 represents the temperature of material point 𝑖 at step 𝑛; 𝑗 denotes a neighboring material
of master material point 𝑖; the discretized heat flux formula of material point 𝑖 and 𝑗 are given as

𝑞𝑖 = −𝑘h [∑ 𝑤(𝜉)(𝛩𝑗 − 𝛩𝑖 ) 𝜉𝑖𝑗 ] 𝑀𝑖−1 (34)

𝑗

𝑞𝑗 = −𝑘h [∑ 𝑤(𝜉)(𝛩𝑙 − 𝛩𝑗 ) 𝜉𝑗𝑙 ] 𝑀𝑗−1 (35)

𝑙

in which the discretized shape tensor of material point 𝑖 and 𝑗, 𝑀𝑖 and 𝑀𝑗 , can be calculated by

𝐽
2
𝑀𝑖 = ∑ 𝑤(𝜉)𝜉𝑖𝑗 (36)
𝑗

𝐽
2 (37)
𝑀𝑗 = ∑ 𝑤(𝜉)𝜉𝑗𝑙
𝑙

Let 𝛩(𝑖, 𝑛) be the exact solution to the PD heat equation, the following equation holds
according to Eq. (26)

𝐽
𝜕𝛩(𝑖, 𝑛) 𝜕𝛩(𝑖, 𝑛) −T 𝜕𝛩(𝑗, 𝑛) −T
𝜌𝑐 = ∑ 𝑤(𝜉) [−𝑘h 𝑀𝑖 − 𝑘h 𝑀𝑗 ] 𝜉 (38)
𝜕𝑡 𝜕𝑥 𝜕𝑥
𝑗

We can further express 𝛩(𝑗, 𝑛) and 𝛩(𝑖, 𝑛 + 1) by the exact solution 𝛩(𝑖, 𝑛) based on the
Taylor expansion

15
 𝜕𝛩(𝑖, 𝑛) 1 𝜕 2 𝛩(𝑖, 𝑛) 2 1 𝜕 3 𝛩(𝑖, 𝑛) 3
𝛩(𝑗, 𝑛) = 𝛩(𝑖, 𝑛) + 𝜉𝑖𝑗 + 𝜉𝑖𝑗 + 𝜉𝑖𝑗 ⋯ (39)
𝜕𝑥 2 𝜕𝑥 2 3! 𝜕𝑥 3

𝜕𝛩(𝑖, 𝑛) 1 𝜕 2 𝛩(𝑖, 𝑛)
𝛩(𝑖, 𝑛 + 1) = 𝛩(𝑖, 𝑛) + ∆𝑡 + (∆𝑡)2 + ⋯ (40)
𝜕𝑡 2 𝜕𝑥 2

Substituting Eqs. (38)-(40) into Eqs. (33)-(35), the truncation error 𝑇𝑖𝑛 of PD heat equation
takes the following form (omitting the spatial terms higher than third order and the temporal terms
higher than second order)

𝐽 𝐽 4 𝐽
1 𝜕 2 𝛩(𝑖, 𝑛) 𝑘ℎ 𝜕 3 𝛩(𝑖, 𝑛) ∑𝑗 𝑤(𝜉) 𝜉𝑖𝑗 𝜕 3 𝛩(𝑗, 𝑛) ∑𝑙 𝑤(𝜉) 𝜉𝑗𝑙4
𝑇𝑖𝑛 = 𝜌𝑐 [ ∆𝑡] + ∑ 𝑤(𝜉) [𝑀𝑖
−T
+ 𝑀𝑗
−T
]𝜉 (41)
2 𝜕𝑥 2 6 𝜕𝑥 3 𝑀𝑖 𝜕𝑥 3 𝑀𝑗
𝑗

The discretized scheme is of one-order accuracy in time and two-order accuracy in space if the
weight function is linearly related to the bond length. Only one-order accuracy in space is achieved
if the unity weight function is employed as will be proven in Eq. (43). One can certainly employ
other higher order integration scheme to achieve higher order accuracy, while it is out of the scope
of this paper since the current scheme is already consistent because the truncation error approaches
zero as ∆𝑡 and 𝜉𝑖𝑗 approaches zero. We want to pay special attention to the effects of horizon and
material point size on the error. Assume a uniform discretization in space leads to |𝜉𝑖𝑗 | = 𝑚∆𝑥,
where 𝑚 represents he 𝑚-th material point away from master material point as shown in Fig. 3.
The shape tensor and truncation error can be rewritten as

𝐿
𝐿(𝐿 + 1)(2𝐿 + 1)
𝑀𝑖 = 𝑀𝑗 = ∑ 2𝑚2 (∆𝑥)2 = (∆𝑥)2 (42)
3
𝑚=1

𝐿
𝑘ℎ 3𝐿2 + 3𝐿 − 1 𝜕 3 𝛩(𝑚+ , 𝑛) 𝜕 3 𝛩(𝑚− , 𝑛)
𝑇𝑖𝑛 = ∑[ − ] 𝑚∆𝑥 (43)
10 𝐿(𝐿 + 1)(2𝐿 + 1) 𝜕𝑥 3 𝜕𝑥 3
𝑚=1

in which we use unity weight function again here for convenience; 𝐿 = 𝐽/2 . Note that the
symmetric property of the material points within horizon is used several times when deriving Eq.
(41) and Eq. (43).

According to Eq. (43), the truncation error is dependent on both the material point size and the
number of neighboring material points, i.e., horizon size. It is clear that the error diminishes as the
material point size approaches zero, while it is not so straightforward to draw a conclusion with
respect to horizon. But it is also not hard to prove that the 𝐿-associated term in Eq. (43) is a

16
monotonically increasing function. Therefore, qualitatively, larger horizon yields more deviations
from the analytical solution. In the following section, we will conduct quantitative parametric
analyses on the effects of horizon and material point size.

Fig. 3. Discretized PD heat equation for 1D rod

3.3 Comparisons between PD and analytical solutions for 1D heat conduction problem

Consider an infinite 1D bar with parameters given as 𝜌 = 1 kg/m3, 𝑐 = 1 J/kg/℃ and 𝑘h = 1

W/m/℃. The initial temperature of the bar 𝛩0 is set to be 0 ℃ and constant temperature 𝛩1 =
100 ℃ is applied on one side of the bar as shown in Fig. 3. The analytical solution of this problem
is given as (Hahn & Özisik, 2012)

𝑥
𝛩(𝑥, 𝑡) = erfc (𝛩1 − 𝛩0 ) + 𝛩0 (44)
𝑘
2√ h 𝑡
( 𝜌𝑐 )

With a uniform mesh and a material point size of 0.02 m, numerical solution is obtained for
the temporal variation of temperature at a material point located one meter away from the heat
boundary. Simulations are performed with varied horizon size and the results are presented in Fig.
4 with comparison to the analytical solution. Evidently, only the numerical results from 𝛿 = ∆𝑥
match well with the analytical solution, while the temperature become much lower than the
analytical solution as horizon increases. This can be well explained by Eq. (32) and Fig. 2, where
a lower dissipation rate is associated with a larger horizon. The discrepancy between PD and
analytical solutions increases nearly exponentially with the horizon size, which is also consistent

17
with the mathematical formulation in Eq. (32). The relative error of the PD heat equation is further
assessed using the relative L1 loss which is calculated by 𝑒1 = (𝑤a − 𝑤n )/𝑤a , where 𝑤a and 𝑤n
represent analytical and numerical solution of temperature at each material point, respectively. As
shown in Fig. 4(b), the numerical solution of temperature is almost always lower than analytical
one due to a smaller dissipation rate of non-local method. Moreover, for 𝛿 = 3∆𝑥 and 𝛿 = 6∆𝑥,
the numerical results exhibit significant oscillations. The peaks and valleys of these two curves
align precisely with the centers of the material points. This phenomenon can be attributed to two
factors: a) the error term described in Eq. (43) increases as the horizon expands, contributing to
the larger deviations from analytical solution observed for 𝛿 = 6∆𝑥; b) for large horizons such as
𝛿 = 3∆𝑥 and 𝛿 = 6∆𝑥, the error term exhibits a highly non-local behavior. In other words, non-
physical effects, such as the particle-skipping heat flux (i.e., heat flow between a pair of points that
jumps intermediate points), are introduced with large non-locality which contribute to the
oscillation in the solution. Note that discretized PD equation is similar to high-order FDM in some
ways, where more than three points are used to approximate the derivatives in both methods.
However, in high-order FDM, it is customary to assign negative coefficients to the non-
neighboring points as means to counterbalance the non-locality introduced by their inclusion,
whereas the weight function in PD is always positive. From this perspective, using PD with large
horizon to model heat conduction process is not recommended. The numerical results for 𝛿 = ∆𝑥
is clearly noise free and the error is generally within 1%.

(a) (b)
Fig. 4. (a) Comparisons of temperature between analytical solution and PD results with different horizons;
and (b) Relative error of PD results.

18
 The error in the simulation may also be reduced by refining the mesh as implied by Eq. (43).
Its effectiveness is examined and presented in Fig. 5(a) for a fixed horizon 𝛿 = 3∆𝑥 with varied
material point size. It is found that while the error is reduced, more severe oscillations are presented
with finer mesh as the error is associated with the position of each material point. The spatial
oscillation of result is undesirable. Hence, merely reducing the material point size may be
insufficient in minimizing the error in a PD model of heat conduction. As a comparison, Fig. 5(b)
shows the relative error with a smaller horizon of 𝛿 = ∆𝑥 with different material point size. Clearly,
a smaller horizon offers a much smoother solution in space. The results imply that reducing the
horizon is more effective and reliable than reducing material point size when minimizing the error
in PD heat conduction model.

(a) (b)
Fig. 5. Relative error of different material point sizes for: (a) 𝛿 = 3∆𝑥 and (b) 𝛿 = ∆𝑥.

Remark 3. The error in the PD heat equation stems from two main sources: a) the slower heat
conduction due to non-locality; and b) the truncation error introduced by the discretization scheme.
Both errors can be mitigated by refining the mesh and reducing the horizon size. However, it is
important to note that the noise in the heat equation, specifically related to particle-skipping heat
flux, can only be effectively reduced by minimizing the horizon size.

4 Multi-horizon peridynamics for multi-physics coupling

In previous sections, we have showed that when solving the peridynamic heat conduction
equation, it is important to adopt an appropriate horizon. A small horizon is effective in mitigating
the numerical oscillations and improving the simulation accuracy for heat conduction problem.
However, as a typical non-local method, PD is primarily designed to capture non-local effects and

19
long-range interactions in material behavior. This includes crack initiation and propagation where
the influence of deformation may extend beyond the immediate vicinity of a material point. In fact,
it is necessary to keep the horizon sufficiently large to capture possible crack branching directions
in dynamic fracture problems (Ha & Bobaru, 2010). Past studies have also shown that good
accuracy can be achieved with the horizon size of 𝛿 = 3Δ𝑥 (Madenci & Oterkus, 2014; Silling &
Askari, 2005) when modeling mechanical material response using PD.
A dilemma arises when developing the coupled thermo-mechanical PD and modeling
thermally induced fracturing (e.g., quenching). On one hand, as mentioned earlier, adopting a small
horizon may not be effective in capturing possible crack branching in different directions. On the
other hand, if a large horizon is adopted, numerical issues such as slower heat conduction rate and
oscillating thermal field can be expected. This is especially true for computations near the
boundary where obvious numerical oscillations are found for temperature field. The errors in
temperature field will induce unphysical thermal expansion or shrinkage which deteriorates the
accuracy of the coupled simulation. While the oscillations can be mitigated by numerical means
such as smoothing or damping (Chang et al., 2022; Gao & Oterkus, 2020; Monaghan, 1994; Yang
et al., 2024), applying those numerical stabilization techniques for heat conduction can lead to an
accelerated dissipation rate of thermal energy and is unpreferred herein. A natural solution for the
coupled thermo-mechanical PD, as introduced below, is to adopt different horizon sizes for
different physics field, to accommodate the distinct characters of the physics fields while
maintaining a decent simulation accuracy.

The idea behind multi-horizon PD is straightforward. As shown in Fig. 6, two different

horizons are defined. One is the classical horizon 𝛿 (hereafter referred to as “horizon”) to capture
the non-local effects in the mechanical field. Another is the thermal horizon 𝛿′, which can be
defined as a constant smaller than the horizon, used to model the local heat conduction. With the
thermal horizon 𝛿′, the thermal neighbor list Ω′𝑥 is established, and Eq. (19) and (21) can be recast
correspondingly as

𝒒𝒙 = −𝑘h [∫ 𝜔⟨‖𝝃‖⟩ 𝛩〈𝒙′ − 𝒙〉 𝝃 d𝑉𝑥 ′ ] 𝑴𝑥−1 (45)

Ω′𝑥

D𝛩
𝜌𝑐 = ∫ [𝜔⟨‖𝝃‖⟩(𝒒𝒙 𝑴−T −T ̇
𝑥 + 𝒒𝒙′ 𝑴𝑥′ )𝝃 + 𝛩𝑩〈𝒙′ − 𝒙〉 ∙ 𝒀〈𝒙′ − 𝒙〉]d𝑉𝑥 ′ + 𝜌𝛩b (46)
D𝑡 ′
Ω𝑥

Note that the mechanical field can still be model by the formulations given in Section 2.1 without
any modifications.

20
 The heat flux equation (Eq. 45) and the heat conduction equation (Eq. 46), including the heat
generated or dissipated by deformation, are solved within the thermal horizon. However, no
kinematical unknowns are updated within thermal horizon even for the thermally induced stress
and deformation. This is illustrated in Fig. 6(a), after solving the thermal field, the shapes of both
the horizon and the thermal horizon remain unchanged, and the positions of all material points
remain unchanged. All the kinematical unknowns (displacement, velocity and acceleration) are
calculated and updated within the horizon as illustrated in Fig. 6(b). The effects of temperature
variation on deformation are considered herein, and the shape of both horizon and thermal horizon
and positions of all material points are updated.

MP after heat transfer

MP after deformation
Horizon
Thermal horizon
Heat conduction Deformed horizon
Deformed thermal horizon

(a)

Deformation

(b)

Fig. 6. Schematic of multi-horizon scheme: (a) Heat conduction process of point 𝑗; and (b) Mechanical
response of point 𝑖.

A staggered computational scheme is adopted to solve the fully coupled thermo-mechanical

PD equations. Neighbor searching is performed at the beginning of the simulation and two
neighbor lists, one based on the horizon and another based on the thermal horizon, are established
and used throughout the simulation. For each step of simulation, the flux and temperature are first
solved with an explicit scheme using forward time difference while other variables remain constant,

21
and the remaining kinematical unknowns are updated with the new temperature using the velocity
Verlet scheme. The computational flow chart is shown in Fig. 7.
It should be noted that the proposed multi-horizon PD approach is distinct from the dual-
horizon PD introduced by Ren et al., (2016, 2017). The dual-horizon PD incorporates variable
horizons at different material points and is suitable for the case where different material point sizes
are used. The proposed multi-horizon PD is for simulation of coupled responses across multiple
physical fields where a distinct horizon can be selected for each physical field.

Start with input files Mechanical part within horizon

Neighbour searching: set up neighbour lists Apply velocity boundary and

based on horizon and thermal horzion calculate by

Apply flux boundary and Apply displacement boundary and

calculate heat flux by Eq. (43)
calculate by

Apply temperature boundary and Calculate by equation of motion

Calculate temperature by Eq. (44)

Apply velocity boundary and

Thermal part within thermal horizon calculate by

Yes

No

End with output files

Fig. 7. Flow chart of computation in the multi-horizon PD.

5 Numerical Examples

5.1 Thermomechanical response of 2D slab under thermal loading

To benchmark the proposed multi-horizon PD, we first simulate the thermomechanical response
of a 2D slab subject to thermal loadings. A 1m by 1 m finite slab with both thermal and mechanical
boundaries is modeled as shown in Fig. 8. The initial temperature of the slab is set to be 0 ℃ and
constant temperature boundary equal to 1 ℃ is applied on the bottom and left boundaries of the
slab. The displacement of bottom and left boundaries is fixed in y and x directions, respectively.
22
The material parameters used are summarized as follows: density 𝜌 = 1 kg/m3 , Young’s modulus
𝐸 = 1 Pa , Poisson’s ratio 𝜇 = 0 .25, liner thermal expansion coefficient 𝛽 = 1 × 10−6 ℃−1 ,
specific hear capacity 𝑐 = 1 J/(kg ∙ ℃), thermal conductivity 𝑘h = 0.1 W/(m ∙ ℃). The material
point size is set as 0.01 m, with 10,000 material points for the slab in total. The horizon is taken to
be 𝛿 = 3Δ𝑥, while the thermal horizon is adopted as 𝛿′ = Δ𝑥. Note that applying the temperature
boundary requires only one layer of material points with the multi-horizon scheme. This serves as
a notable advantage, particularly for large-scale 3D models, as it allows for significant reduction
in the number of material points required for the boundary. Since there is no available analytical
solution for this particular transient problem, the commercial software COMSOL Multiphysics
based on finite element method (FEM) is employed to benchmark the proposed method.

Fig. 8. Mesh and boundary conditions of a 2D slab.

The temperature and displacement distribution in the slab after 0.5s simulation obtained from
FEM and multi-horizon PD are shown in Fig. 9. Evidently, the prediction of temperature and
displacement distribution by multi-horizon PD is consistent with that by FEM. A further
quantitative comparison of the temporal evolutions of temperature and displacement magnitude at
point (0.5,0.5) is shown in Fig. 10. Again, good agreements are observed between FEM and multi-
horizon PD. For comparison purpose, we also included the results from two additional PD models
where a single horizon is adopted for both mechanical and thermal fields. Observably, when a
horizon of 𝛿 = Δ𝑥 is used, reasonable results for temperature are obtained whereas the
displacement is far from satisfactory. When a horizon of 𝛿 = 3Δ𝑥 is adopted, mechanical
responses are well captured but the temperature is not in good agreement with the FEM results.
The results justify the necessity of adopting multi horizons for different physical fields.
Furthermore, we examine the temperature contours obtained from the multi-horizon

23
thermomechanical PD model with comparison to the original PD model (i.e., where a single
horizon is used) as shown in Fig. 11. The results from original PD model exhibit significant
oscillations due to the reasons described in Section 3. In contrast, the temperature contours from
the multi-horizon PD model are clearly smooth and noise-free. Hence, the results prove the
capability of the multi-horizon PD approach in accurately modeling both the mechanical and
thermal fields by utilization of distinct horizons to capture non-local and local responses
effectively.

Fig. 9. Results obtained from FEM for: (a) temperature; (b) displacement in x direction; (c) displacement
in y direction; and results obtained by multi-horizon PD for: (d) temperature; (e) displacement in x direction;
(f) displacement in y direction.

24
 0.25 18
Original TM PD with d = 3 Dx Original TM PD with d = 3Dx
Original TM PD with d = Dx 16
Original TM PD with d = Dx

Displacement, u and v: ×10-8 m

0.20 Multi-horizon PD Multi-horizon PD
FEM
14
FEM
Temperature, Q : °C

12
0.15
10

0.10 8

6
0.05 4

2
0.00
0

0.0 0.1 0.2 0.3 0.4 0.5 0.0 0.1 0.2 0.3 0.4 0.5
Time, t : s Time, t : s

(a) (b)
Fig. 10. Evolutions of: (a) temperature and (b) displacement magnitude at point (0.5,0.5) (u and v represent
displacement along x and y directions, respectively).

(a) (b)
Fig. 11. Temperature contours of multi-horizon PD and original PD at 𝑡 = 0.5 s.

25
5.2 Thermomechanical response of 3D beam under thermal loading

A thermomechanical model for a 3D beam has been created to further test the performance of the
proposed multi-horizon PD as shown in Fig. 12. One end of the beam is subjected to a constant
temperature boundary condition where the temperature is kept at 1 ℃ and the other end of the
beam is fixed. The beam has dimensions of 1m×1m×4m and is discretized into 62,500 uniform
material points, with each point having a size of 4 cm. The temperature boundary is applied by one
additional layer of material points (625 material points) outside the left surface of the beam. All
the material parameters used in the 3D case are the same as those used in the 2D case in the last
section except for thermal conductivity 𝑘h , which is set to 1 W/(m ∙ ℃). Again, the horizon is
taken to be 𝛿 = 3Δ𝑥, while the thermal horizon is adopted as 𝛿′ = Δ𝑥.

(0,0,0)

Fig. 12. Mesh and boundary conditions of a 3D beam.

The obtained results are presented in Fig. 13. Significant temperature noise can be observed
both along the length and inside the beam when employing the original thermomechanical PD.
Further examination of the temperature distribution at various cross-sections is given in Fig. 14(a)-
(d). The observed temperature noise, primarily resulting from the truncation error and excessive
number of neighbors, appears to manifest as a random pattern. With the introduction of the multi-
horizon scheme, the oscillation is significantly mitigated and much smoother temperature profiles
both along and inside the beam are obtained as shown in Fig. 13(d)-(f) and Fig. 14(e)-(f).

26
Fig. 13. Temperature colormap obtained from original TM PD at (a) t = 0.1 s; (b) t = 0.5 s; (c) t = 1 s; and
from multi-horizon PD at (d) t = 0.1 s; (e) t = 0.5 s; (f) t = 1 s.

Quantitative comparisons with FEM results at the point (1,0.25,0.25) are depicted in Fig. 15.
The FEM results are obtained from COMSOL Multiphysics. The multi-horizon PD demonstrates
superior performance compared to the original thermomechanical PD method, regardless of
whether a horizon size of 𝛿 = 3Δ𝑥 or 𝛿 = Δ𝑥 is used. Moreover, the multi-horizon method seems
to offer even smoother results for in the displacements in x direction than FEM as can be seen in
Fig. 15(b).

Fig. 14. Snapshots of temperature distribution at different cross-sections obtained from original TM PD: (a)
x = 0.2; (b) x = 0.5; (c) x = 1.0; (d) x = 1.7; and from multi-horizon PD (e) x = 0.2 ; (f) x = 0.5 ; (g) x = 1.0;
(h) x = 1.7.
27
Fig. 15. Evolutions of: (a) temperature and (b) displacements at point (1, 0.25, 0.25) (u and v represent
displacement along x and y directions, respectively).

5.3 Fully coupled responses of plate subjected to combined force and temperature

Although based on the fully coupled multi-horizon PD method, the above two cases only involve
thermally induced thermal and mechanical responses. The case in this section will focus on a plate
subjected to coupled thermal and mechanical loads. As shown in Fig. 16, the left boundary of the
plate is subjected to a 1 Pa normal pressure and a constant temperature equal to 1 ℃, while the
right boundary is fixed along the x direction. The geometry of the plate and discretization are the
same as the 2D case in Section 5.1. The material parameters used are summarized as follows:
density 𝜌 = 1 kg/m3 , Young’s modulus 𝐸 = 1 Pa , Poisson’s ratio 𝜇 = 0.25 , liner thermal
expansion coefficient 𝛽 = 1 × 10−6 ℃−1 , specific heat capacity 𝑐 = 1 J/(kg ∙ ℃) , thermal
conductivity 𝑘h = 0.5 𝑊/(m ∙ ℃). Again, the horizon is taken to be 𝛿 = 3Δ𝑥, while the thermal
horizon is adopted as 𝛿′ = Δ𝑥.

The same problem was analysed by Hosseini-Tehrani & Eslami (2000) using boundary
element method (BEM), and therefore their results are used to verify the proposed multi-horizon
scheme. Results of the simulation are presented in Fig. 17 for the evolution of temperature and
magnitude of displacement for a point at (0.2, 0.5). The results obtained from the multi-horizon
PD are in good agreement with the BEM results with respect to both temperature and displacement,
confirming the capacity of the proposed method in simulating fully coupled thermo-mechanical
response of material.

28
 Fig. 16. Model of a 2D plate subject to combined thermal and mechanical loadings.

Fig. 17. Evolutions of: (a) temperature and (b) displacement magnitude at point (0.2,0.5).

5.4 Thermally induced fracturing in granite

To further demonstrate the capability of the proposed multi-horizon PD in modeling the fracturing
process induced by various physical fields beyond mechanical loading, the thermally induced
fracturing in an unconfined granite is modeled. The model setup follows an experiment conducted
on Lac Du Bonnet (LDB) granite (Jansen et al., 1993), which has a density of 2650 kg/m3, an
elastic modulus of 67 GPa, a Poisson’s ratio of 0.33, a thermal expansion coefficient of 3.5×10-6
1/°C, a specific heat of 1015 J/kg/°C, and a thermal conductivity of 3.5 W/m/°C. The rock
specimen is a 15-cm cube with a 1-cm-diameter vertical borehole drilled at the center of the
specimen, which can be well simplified into a 2D plane strain numerical model as shown in Fig.
18(c). The numerical model is discretized into 22,384 material points, with each material point
size equal to 1 mm. Note that the mesh is not totally symmetric owing to the fact that the granite
in experiment is not isotropic and homogenous. The granite is modeled by the multi-horizon PD

29
with horizon set as three times the material point size and thermal horizon set as 1.2 times the
material point size, along with the critical stretch damage model and tension failure model as
described in Section 2.3. The critical energy release rate and tensile strength of LDB granite are
adopted as 70 J/m2 (Wang et al., 2018) and 9 MPa (Martin, 1994), respectively. Note that the
critical energy release rate and, consequently, the critical stretch can both be temperature
dependent. For example, the fracture toughness of granite was found to decrease with rising
temperature (Feng et al., 2019; Ge et al., 2021). Nonetheless, such effect is apparent only after a
long duration heating process. In the present study, all the material properties are assumed to be
temperature-independent constants. When necessary, the temperature-dependency of material
fracture strength can be further considered by incorporating a relationship between temperature
and critical energy release rate into the damage model presented in Section 2.3. The initial
temperature and temperature of the four outer boundaries of the model are set as 20 °C, while
thermal loading with a rate of 0.05 °C/s is applied on the inner circular boundary. Simulation is
performed to 2,000 s with time step of 2×10-7 s.

20 °C

20+0.05t °C

(a) (b) (c)

Fig. 18. (a) Experiment setup (Jansen et al., 1993); (b) Sketches of major fractures in the experiment
(Jansen et al., 1993); and (c) Numerical model.

30
 Fig. 19. Evolutions of temperature, displacement in x direction, damage and major principal stress at
different times.

Fig. 19 shows the distribution of several crucial variables, including temperature, displacement
in x direction, damage and major principal stress at different times. As the temperature applied to
the borehole increases, heat is conducted from the inner boundary towards the outer boundary.
This results in the formation of compressive stress near the borehole, while tensile stress is induced
in certain areas of the free surface. These stresses result from temperature gradient between inner
and outer part of the model and relatively high transverse thermal expansion of inner part with
respect to outer part. When the borehole is heated to 102.88 °C at about 1657.6 s, the tensile stress
exceeds the tensile strength of the granite and therefore fracture is initiated from the outer surface
first. Extensive tests conducted by Jackson et al. (1989) reported that the critical fracture
temperature for LDB granite falls in the range of 80°C-125°C, which concurs with the numerical
results. Once the initial crack forms, the tensile stress becomes localized at the crack front, which
drives the crack to propagate continuously, while the tensile stress at other positions tends to
dissipate. Additionally, a significant jump in displacement is observed along the two sides of the

31
crack. This displacement jump is a result of the release of stored elastic energy during crack
propagation. With further increase in temperature, a second crack is initiated at the lower surface
and the two cracks finally propagate nearly parallelly to the edge of borehole. The fracturing
pattern is consistent with that reported by literatures (Jackson et al., 1989; Jansen et al., 1993), and
the morphology of crack as shown in Fig. 19 match well with the test results given in Fig. 18(b).
The presence of the crack hinders the conduction of heat but does not completely interrupt it. As
depicted in the top-right figure of Fig. 19, heat is still able to transfer outwards after the penetration
cracks have formed. However, the temperature distributions on different sides of the crack are no
longer symmetric due to the hindrance caused by the crack. The asymmetry in temperature
distribution is a direct consequence of the presence and effects of the crack on the heat transfer
process. This example well demonstrates the capability of the multi-horizon PD in modeling
evolving discontinuities subject to thermal loading.

6 Conclusions and discussion

This paper presents a new fully coupled thermomechanical state-based PD model that combines
non-local operators to solve the heat conduction equation with a thermoelastic PD solid model.
The dispersion analysis based on spectral approach and error analysis based on Taylor expansion
reveal that using large horizons induces more error and noise, resulting in a decrease in the
conduction rate. To address this issue, this paper proposes an innovative multi-horizon scheme
that uses different horizons for different physical fields. For thermomechanical cases, a smaller
horizon for thermal field and a larger horizon for the mechanical field are adopted to ensure the
accuracy of heat conduction simulation while maintaining the capability of capturing non-local
effects in mechanical responses. The numerical results show that the multi-horizon scheme offers
noise-free and more accurate solutions. The scheme allows for the adoption of distinct horizon
sizes that are most suitable for capturing the specific characteristics and phenomena associated
with a particular physical field and serves a useful tool when coupling different physical fields in
PD. Although the presented approach focuses on the TM PD, the generality of this paper is not
lost owing to the complete analogy between different physical fields within solid (Mitchell & Soga,
2005). The presented approach can indeed be used to derive more sophisticated multi-physics PD-
based computational framework in the future, including but not limited to coupled
hydromechanical, electromechanical, and mechanical-chemical processes.

Due to limited length of this paper, it has not been possible to exhaustively explore all aspects
of the multi-horizon PD. One potential issue that requires further investigation is the mesh
dependency of the model. The presented numerical examples utilize a uniform mesh for optimal
performance. However, when using an irregular discretization pattern, instability issues may arise,

32
particularly when dealing with physical fields other than the mechanical field. This might be
partially attributed to the poor evaluation of derivatives from integration over irregularly
distributed material points. Further efforts are needed to study the influence of mesh on numerical
stability. Additionally, in scenarios involving coupled heat contact and cracking problems across
different materials or phases, the development of a multi-resolution scheme (Yao et al., 2023) may
become necessary. However, this topic is beyond the scope of the current study and will be
addressed separately in future research. Another concern is the computational cost for the pure
PD-based framework, particularly for 3D simulations with a large material stiffness and a small
thermal conductivity. In such scenarios, a small time step may be necessary which would trigger
overwhelming computational cost. Adopting an implicit scheme and/or GPU-based parallel
computing may be beneficial. In addition, coupling PD with local methods such as FEM (Ni et al.,
2021, 2022) may also be advantageous for saving computational cost by splitting computations of
different physical fields between PD and other methods.

Acknowledgments
The study was financially supported by the National Science Foundation of China under project
11972030, Research Grants Council of Hong Kong (through GRF project 16208720, CRF project
C7082-22G, TRS project T22-606/23-R, and HKUST SSTSP FP907), and Hetao Shenzhen-Hong
Kong Science and Technology Innovation Cooperation Zone (HZQBKCZYB2020083). Fan Zhu
acknowledges support from JSPS KAKENHI Grant Number 23K13403. Changyi Yang
acknowledges financial support from the Hong Kong Ph.D. Fellowship Scheme funded by
Research Grants Council of Hong Kong and would like to express gratitude to Prof. Erdogan
Madenci from the University of Arizona, and Prof. Heng Zhang from Hohai University for the
beneficial discussions.

References

Agwai, A. G. (2011). A Peridynamic Approach for Coupled Fields.

https://repository.arizona.edu/handle/10150/204892
Bažant, Z. P., Luo, W., Chau, V. T., & Bessa, M. A. (2016). Wave Dispersion and Basic Concepts
of Peridynamics Compared to Classical Nonlocal Damage Models. Journal of Applied
Mechanics, 83(11), 111004. https://doi.org/10.1115/1.4034319
Bazazzadeh, S., Mossaiby, F., & Shojaei, A. (2020). An adaptive thermo-mechanical peridynamic
model for fracture analysis in ceramics. Engineering Fracture Mechanics, 223, 106708.
https://doi.org/10.1016/j.engfracmech.2019.106708

33
Bergel, G. L., & Li, S. (2016). The total and updated lagrangian formulations of state-based
peridynamics. Computational Mechanics, 58(2), 351–370. https://doi.org/10.1007/s00466-
016-1297-8
Bessa, M. A., Foster, J. T., Belytschko, T., & Liu, W. K. (2014). A meshfree unification:
Reproducing kernel peridynamics. Computational Mechanics, 53(6), 1251–1264.
https://doi.org/10.1007/s00466-013-0969-x
Buckman, S., Morris, R. H., & Bourman, R. P. (2021). Fire-induced rock spalling as a mechanism
of weathering responsible for flared slope and inselberg development. Nature
Communications, 12(1), Article 1. https://doi.org/10.1038/s41467-021-22451-2
Butt, S. N., Timothy, J. J., & Meschke, G. (2017). Wave dispersion and propagation in state-based
peridynamics. Computational Mechanics, 60(5), 725–738. https://doi.org/10.1007/s00466-
017-1439-7
Chang, H., Chen, A., Kareem, A., Hu, L., & Ma, R. (2022). Peridynamic differential operator-
based Eulerian particle method for 2D internal flows. Computer Methods in Applied
Mechanics and Engineering, 392, 114568. https://doi.org/10.1016/j.cma.2021.114568
Chen, W., Gu, X., Zhang, Q., & Xia, X. (2021). A refined thermo-mechanical fully coupled
peridynamics with application to concrete cracking. Engineering Fracture Mechanics, 242,
107463. https://doi.org/10.1016/j.engfracmech.2020.107463
Chen, Z., & Bobaru, F. (2015). Peridynamic modeling of pitting corrosion damage. Journal of the
Mechanics and Physics of Solids, 78, 352–381. https://doi.org/10.1016/j.jmps.2015.02.015
Feng, G., Kang, Y., & Wang, X. (2019). Fracture failure of granite after varied durations of thermal
treatment: An experimental study. Royal Society Open Science, 6(7), 190144.
https://doi.org/10.1098/rsos.190144
Gao, Y., & Oterkus, S. (2019). Fully coupled thermomechanical analysis of laminated composites
by using ordinary state based peridynamic theory. Composite Structures, 207, 397–424.
https://doi.org/10.1016/j.compstruct.2018.09.034
Gao, Y., & Oterkus, S. (2020). Fluid-elastic structure interaction simulation by using ordinary
state-based peridynamics and peridynamic differential operator. Engineering Analysis with
Boundary Elements, 121, 126–142. https://doi.org/10.1016/j.enganabound.2020.09.012
Ge, Z., Sun, Q., Yang, T., Luo, T., Jia, H., & Yang, D. (2021). Effect of high temperature on mode-
I fracture toughness of granite subjected to liquid nitrogen cooling. Engineering Fracture
Mechanics, 252, 107834. https://doi.org/10.1016/j.engfracmech.2021.107834

34
Gu, X., Zhang, Q., Huang, D., & Yv, Y. (2016). Wave dispersion analysis and simulation method
for concrete SHPB test in peridynamics. Engineering Fracture Mechanics, 160, 124–137.
https://doi.org/10.1016/j.engfracmech.2016.04.005
Ha, Y. D., & Bobaru, F. (2010). Studies of dynamic crack propagation and crack branching with
peridynamics. International Journal of Fracture, 162(1), 229–244.
https://doi.org/10.1007/s10704-010-9442-4
Hahn & Özisik. (2012). Heat conduction.
Hosseini-Tehrani, P., & Eslami, M. R. (2000). BEM analysis of thermal and mechanical shock in
a two-dimensional finite domain considering coupled thermoelasticity. Engineering
Analysis with Boundary Elements, 24(3), 249–257. https://doi.org/10.1016/S0955-
7997(99)00063-6
Jackson, R., Lau, J. S. O., & Annor, A. (1989). Mechanical, thermo-mechanical and joint
properties of rock samples from the site of AECL’s Underground Research Laboratory,
Lac du Bonnet, Manitoba. Materials: From Theory to Practice, Proceedings of the 42nd
Canadian Geotechnical Conference, Winnipeg. Canadian Geotechnical Society, Toronto,
41–49.
Jansen, D. P., Carlson, S. R., Young, R. P., & Hutchins, D. A. (1993). Ultrasonic imaging and
acoustic emission monitoring of thermally induced microcracks in Lac du Bonnet granite.
Journal of Geophysical Research: Solid Earth, 98(B12), 22231–22243.
https://doi.org/10.1029/93JB01816
Kilic, B., & Madenci, E. (2010). Peridynamic Theory for Thermomechanical Analysis. IEEE
Transactions on Advanced Packaging, 33(1), 97–105.
https://doi.org/10.1109/TADVP.2009.2029079
Kulkarni, S., & Tabarraei, A. (2018). An analytical study of wave propagation in a peridynamic
bar with nonuniform discretization. Engineering Fracture Mechanics, 190, 347–366.
https://doi.org/10.1016/j.engfracmech.2017.12.019
Le, Q. v., Chan, W. k., & Schwartz, J. (2014). A two-dimensional ordinary, state-based
peridynamic model for linearly elastic solids. International Journal for Numerical Methods
in Engineering, 98(8), 547–561. https://doi.org/10.1002/nme.4642
Liu, X., Yu, P., Zheng, B., Oterkus, E., He, X., & Lu, C. (2024). Prediction of graphene’s
mechanical and fracture properties via peridynamics. International Journal of Mechanical
Sciences, 266, 108914. https://doi.org/10.1016/j.ijmecsci.2023.108914
Madenci, E., Barut, A., & Dorduncu, M. (2019). Peridynamic Differential Operator for Numerical
Analysis. Springer International Publishing. https://doi.org/10.1007/978-3-030-02647-9

35
Madenci, E., & Oterkus, E. (2014). Peridynamic Theory and Its Applications. Springer New York.
https://doi.org/10.1007/978-1-4614-8465-3
Martin, C. D. (1994). The strength of massive Lac du Bonnet granite around underground
openings. National Library of Canada = Bibliothèque nationale du Canada.
Menon, S., & Song, X. (2021). A stabilized computational nonlocal poromechanics model for
dynamic analysis of saturated porous media. International Journal for Numerical Methods
in Engineering, 122(20), 5512–5539. https://doi.org/10.1002/nme.6762
Menon, S., & Song, X. (2022). Computational multiphase periporomechanics for unguided
cracking in unsaturated porous media. International Journal for Numerical Methods in
Engineering, 123(12), 2837–2871. https://doi.org/10.1002/nme.6961
Menon, S., & Song, X. (2023). Computational coupled large-deformation periporomechanics for
dynamic failure and fracturing in variably saturated porous media. International Journal
for Numerical Methods in Engineering, 124(1), 80–118. https://doi.org/10.1002/nme.7109
Mikata, Y. (2012). Analytical solutions of peristatic and peridynamic problems for a 1D infinite
rod. International Journal of Solids and Structures, 49(21), 2887–2897.
https://doi.org/10.1016/j.ijsolstr.2012.02.012
Mitchell, J. K., & Soga, K. (n.d.). Fundamentals of Soil Behavior. 2005.
Monaghan, J. J. (1994). Simulating Free Surface Flows with SPH. Journal of Computational
Physics, 110(2), 399–406. https://doi.org/10.1006/jcph.1994.1034
Ni, T., Fan, X., Zhang, J., Zaccariotto, M., Galvanetto, U., & Schrefler, B. A. (2023). A
Peridynamic-enhanced finite element method for Thermo–Hydro–Mechanical coupled
problems in saturated porous media involving cracks. Computer Methods in Applied
Mechanics and Engineering, 417, 116376. https://doi.org/10.1016/j.cma.2023.116376
Ni, T., Sanavia, L., Zaccariotto, M., Galvanetto, U., & Schrefler, B. A. (2022). Fracturing dry and
saturated porous media, Peridynamics and dispersion. Computers and Geotechnics, 151,
104990. https://doi.org/10.1016/j.compgeo.2022.104990
Ni, T., Zaccariotto, M., Zhu, Q.-Z., & Galvanetto, U. (2021). Coupling of FEM and ordinary state-
based peridynamics for brittle failure analysis in 3D. Mechanics of Advanced Materials
and Structures, 28(9), 875–890. https://doi.org/10.1080/15376494.2019.1602237
Oterkus, S., Madenci, E., & Agwai, A. (2014a). Fully coupled peridynamic thermomechanics.
Journal of the Mechanics and Physics of Solids, 64, 1–23.
https://doi.org/10.1016/j.jmps.2013.10.011

36
Oterkus, S., Madenci, E., & Agwai, A. (2014b). Peridynamic thermal diffusion. Journal of
Computational Physics, 265, 71–96. https://doi.org/10.1016/j.jcp.2014.01.027
Oterkus, S., Madenci, E., & Oterkus, E. (2017). Fully coupled poroelastic peridynamic formulation
for fluid-filled fractures. Engineering Geology, 225, 19–28.
https://doi.org/10.1016/j.enggeo.2017.02.001
Rabczuk, T., Ren, H., & Zhuang, X. (2019). A Nonlocal Operator Method for Partial Differential
Equations with Application to Electromagnetic Waveguide Problem. Computers,
Materials & Continua, 59(1), 31–55. https://doi.org/10.32604/cmc.2019.04567
Ren, H., Zhuang, X., Cai, Y., & Rabczuk, T. (2016). Dual-horizon peridynamics. International
Journal for Numerical Methods in Engineering, 108(12), 1451–1476.
https://doi.org/10.1002/nme.5257
Ren, H., Zhuang, X., & Rabczuk, T. (2020a). A higher order nonlocal operator method for solving
partial differential equations. Computer Methods in Applied Mechanics and Engineering,
367, 113132. https://doi.org/10.1016/j.cma.2020.113132
Ren, H., Zhuang, X., & Rabczuk, T. (2020b). A nonlocal operator method for solving partial
differential equations. Computer Methods in Applied Mechanics and Engineering, 358,
112621. https://doi.org/10.1016/j.cma.2019.112621
Ren, H., Zhuang, X., & Timon, R. (2017). Dual-horizon peridynamics: A stable solution to varying
horizons. Computer Methods in Applied Mechanics and Engineering, 318, 762–782.
https://doi.org/10.1016/j.cma.2016.12.031
Ruan, H., Rezaei, S., Yang, Y., Gross, D., & Xu, B.-X. (2023). A thermo-mechanical phase-field
fracture model: Application to hot cracking simulations in additive manufacturing. Journal
of the Mechanics and Physics of Solids, 172, 105169.
https://doi.org/10.1016/j.jmps.2022.105169
Shi, K., Zhu, F., & Zhao, J. (2022). Multi-scale analysis of shear behaviour of crushable granular
sand under general stress conditions. Géotechnique, 1–18.
https://doi.org/10.1680/jgeot.21.00412
Silling, S. A. (2000). Reformulation of elasticity theory for discontinuities and long-range forces.
Journal of the Mechanics and Physics of Solids, 48(1), 175–209.
https://doi.org/10.1016/S0022-5096(99)00029-0
Silling, S. A. (2016). Solitary waves in a peridynamic elastic solid. Journal of the Mechanics and
Physics of Solids, 96, 121–132. https://doi.org/10.1016/j.jmps.2016.06.001

37
Silling, S. A., & Askari, E. (2005). A meshfree method based on the peridynamic model of solid
mechanics. Computers & Structures, 83(17), 1526–1535.
https://doi.org/10.1016/j.compstruc.2004.11.026
Silling, S. A., Epton, M., Weckner, O., Xu, J., & Askari, E. (2007). Peridynamic States and
Constitutive Modeling. Journal of Elasticity, 88(2), 151–184.
https://doi.org/10.1007/s10659-007-9125-1
Song, X., & Silling, S. A. (2020). On the peridynamic effective force state and multiphase
constitutive correspondence principle. Journal of the Mechanics and Physics of Solids, 145,
104161. https://doi.org/10.1016/j.jmps.2020.104161
Song, Y., Li, S., & Li, Y. (2022). Peridynamic modeling and simulation of thermo-mechanical
fracture in inhomogeneous ice. Engineering with Computers.
https://doi.org/10.1007/s00366-022-01616-7
Song, Y., Li, S., & Zhang, S. (2021). Peridynamic modeling and simulation of thermo-mechanical
de-icing process with modified ice failure criterion. Defence Technology, 17(1), 15–35.
https://doi.org/10.1016/j.dt.2020.04.001
Sun, W., Fish, J., & Guo, C. (2022). Parallel PD-FEM simulation of dynamic fluid-driven fracture
branching in saturated porous media. Engineering Fracture Mechanics, 274, 108782.
https://doi.org/10.1016/j.engfracmech.2022.108782
Sun, W., Fish, J., & Lin, P. (2022). Numerical simulation of fluid-driven fracturing in orthotropic
poroelastic media based on a peridynamics-finite element coupling approach. International
Journal of Rock Mechanics and Mining Sciences, 158, 105199.
https://doi.org/10.1016/j.ijrmms.2022.105199
Sun, W., Susmel, L., & Xie, P. (2023). Thermal Fracturing in Orthotropic Rocks with
Superposition-Based Coupling of PD and FEM. Rock Mechanics and Rock Engineering,
56(3), 2395–2416. https://doi.org/10.1007/s00603-022-03164-4
Tu, Q., & Li, S. (2017). An updated Lagrangian particle hydrodynamics (ULPH) for Newtonian
fluids. Journal of Computational Physics, 348, 493–513.
https://doi.org/10.1016/j.jcp.2017.07.031
Wan, J., Chen, Z., Chu, X., & Liu, H. (2020). Dependency of single-particle crushing patterns on
discretization using peridynamics. Powder Technology, 366, 689–700.
https://doi.org/10.1016/j.powtec.2020.03.021
Wang, H. F. (2017). Theory of Linear Poroelasticity with Applications to Geomechanics and
Hydrogeology. In Theory of Linear Poroelasticity with Applications to Geomechanics and
Hydrogeology. Princeton University Press. https://doi.org/10.1515/9781400885688

38
Wang, H., Oterkus, E., & Oterkus, S. (2018). Peridynamic modelling of fracture in marine lithium-
ion batteries. Ocean Engineering, 151, 257–267.
https://doi.org/10.1016/j.oceaneng.2018.01.049
Wang, Y., Zhou, X., & Kou, M. (2018). A coupled thermo-mechanical bond-based peridynamics
for simulating thermal cracking in rocks. International Journal of Fracture, 211(1), 13–42.
https://doi.org/10.1007/s10704-018-0273-z
Wang, Y., Zhou, X., & Kou, M. (2019). An improved coupled thermo-mechanic bond-based
peridynamic model for cracking behaviors in brittle solids subjected to thermal shocks.
European Journal of Mechanics - A/Solids, 73, 282–305.
https://doi.org/10.1016/j.euromechsol.2018.09.007
Wu, P., & Chen, Z. (2023). Peridynamic electromechanical modeling of damaging and cracking
in conductive composites: A stochastically homogenized approach. Composite Structures,
305, 116528. https://doi.org/10.1016/j.compstruct.2022.116528
Xue, Y., Liu, S., Chai, J., Liu, J., Ranjith, P. G., Cai, C., Gao, F., & Bai, X. (2023). Effect of water-
cooling shock on fracture initiation and morphology of high-temperature granite:
Application of hydraulic fracturing to enhanced geothermal systems. Applied Energy, 337,
120858. https://doi.org/10.1016/j.apenergy.2023.120858
Yang, C., Li, L., & Li, J. (2020). Service life of reinforced concrete seawalls suffering from
chloride attack: Theoretical modelling and analysis. Construction and Building Materials,
263, 120172. https://doi.org/10.1016/j.conbuildmat.2020.120172
Yang, C., Zhu, F., & Zhao, J. (2024). Coupled total- and semi-Lagrangian peridynamics for
modelling fluid-driven fracturing in solids. Computer Methods in Applied Mechanics and
Engineering, 419, 116580. https://doi.org/10.1016/j.cma.2023.116580
Yao, X., Chen, D., Wu, L., & Huang, D. (2023). A multi-resolution DFPM-PD model for efficient
solution of FSI problems with structural deformation and failure. Engineering Analysis
with Boundary Elements, 157, 424–440.
https://doi.org/10.1016/j.enganabound.2023.09.023
Zhang, H., & Qiao, P. (2018). An extended state-based peridynamic model for damage growth
prediction of bimaterial structures under thermomechanical loading. Engineering Fracture
Mechanics, 189, 81–97. https://doi.org/10.1016/j.engfracmech.2017.09.023
Zhang, H., & Qiao, P. (2020). A two-dimensional ordinary state-based peridynamic model for
elastic and fracture analysis. Engineering Fracture Mechanics, 232, 107040.
https://doi.org/10.1016/j.engfracmech.2020.107040

39
Zhang, Y., Yu, S., & Deng, H. (2023). Peridynamic model of deformation and failure for rock
material under the coupling effect of multi-physical fields. Theoretical and Applied
Fracture Mechanics, 125, 103912. https://doi.org/10.1016/j.tafmec.2023.103912
Zhu, F., & Zhao, J. (2019a). A peridynamic investigation on crushing of sand particles.
Géotechnique, 69(6), 526–540. https://doi.org/10.1680/jgeot.17.P.274
Zhu, F., & Zhao, J. (2019b). Modeling continuous grain crushing in granular media: A hybrid
peridynamics and physics engine approach. Computer Methods in Applied Mechanics and
Engineering, 348, 334–355. https://doi.org/10.1016/j.cma.2019.01.017

Author statement

Changyi Yang: Conceptualization, Methodology, Software, Data curation, Writing- Original

draft preparation
Fan Zhu: Software, Validation, Visualization, Writing- Reviewing and Editing.
Jidong Zhao: Conceptualization, Supervision, Investigation, Validation, Writing- Reviewing and
Editing, Funding acquisition.

Declaration of interests

☒ The authors declare that they have no known competing financial interests or personal
relationships that could have appeared to influence the work reported in this paper.

☐The authors declare the following financial interests/personal relationships which may be considered
as potential competing interests:

40